1. Dynamic embedding strategy of VQ-based information hiding approach
Source : Journal of Visual Communication and Image Representation, vol. 59, pp , February 2019
Authors : Cheng-Ta Huang, Li-Chiun Lin, Cheng-Hsing Yang and Shiuh-Jeng Wang
Speaker : Chia-Shuo Shih
Date : 2019/02/28 1

2. Outline Introduction Related work Proposed method Experimental results
Conclusions 2

3. Introduction(1/2)-data hiding
Secret bits
Original image
Stego image 3

4. Introduction(2/2)-Compression
512 * 512 = (pixels)
* 8 = (bits)
Compression ratio(CR) = 0.5
(bits) (bits)
Original image
Compressed image
lossy compression
lossless compression
Data hiding
Improve the image quality 4

5. Related work(1/3)-Vector Quantization
VQ is a lossy image coding technique.
VQ consists of three procedures:
Codebook design
Generate a set of representative codewords.
The LBG algorithm is the most commonly used method.
Image Encoding
Image Decoding
[34] Y. Linde, A. Buzo, R. Gray, “Algorithm for Vector Quantizer Design”, IEEE Transactions on Communications, vol. 28, no. 1, pp: 84-95, 1980. 5

6. Related work(2/3)-Vector Quantization
Image Encoding
Partition the image into non-overlapped image block
Find the closest codeword in the codebook for each image block x
The index of the closest codeword of x is recorded.
Image Decoding
Reconstruct each block by the codeword in the codebook of its index. 6

7. Related work(3/3)-Vector Quantization
Codebook 7

8. Proposed method(1/9) 𝐶 𝑖 𝑊 𝑖 𝑊 𝑖 ′′′ 8 support take out LSB VQ
Secret bits
Index block 8

9. Proposed method(2/9) 4 4 … 65 77 78 90 84 66 70 72 79 80 83 75 69 67
Original image 9

10. Proposed method(3/9) 𝐹 𝑖 10 Codebook size = 512 I = 65 0 0100 0001 277 78 90 84 66 70 72 79 80 83 75 69 67 60 70 64 80 78 75 79 74 69 77 76 VQ
𝑊 𝑖 = {15,17,16,20,19,20,18,18,19,37,34,39,35,38,40,38}
Image block
Index
𝐶 𝑖 = {16,19,19,22,21,16,17,18,19,40,41,37,33,34,33,36}
Codebook Index
𝐹 𝑖
65 =
77 = .
79 =
67 =
72 =
32 =
38 = .
39 =
67 =
72 =
16 =
19 = .
33 =
36 =
{65,77,78,90,84,66,70,72,79,80,83,75,66,69,67,72}
{32,38,39,45,42,33,35,36,39,80,83,75,66,69,67,72} 10

11. Proposed method(4/9)
T = 3 S = [ ]
𝐶 𝑖 = {16,19,19,22,21,16,17,18,19,40,41,37,33,34,33,36} 𝐹 𝑖 = { 0 }
𝑊 𝑖 = {15,17,16,20,19,20,18,18,19,37,34,39,35,38,40,38} 𝑊 𝑖 ′ = { 19 }
𝑊 𝑖 ′ ＝ 𝑊 𝑖 +𝑏, 𝑖𝑓 𝐶 𝑖 > 𝑊 𝑖 𝑊 𝑖 −𝑏, 𝑖𝑓 𝐶 𝑖 < 𝑊 𝑖 𝑊 𝑖 ,𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
𝐶 1 = 16
𝑊 1 = 15
𝐹 1 = 0
S = [10]
S = [100]
S = [1001]
S = [1]
𝑊 1 ′ = 19
𝑊 1 + S ≦ 𝐶 1 + T ≦
𝑊 1 + S ≦ 𝐶 1 + T ≦
𝑊 1 + S ≦ 𝐶 1 + T ≦
𝑊 1 + S ≦ 𝐶 1 + T ≦ 11

12. Proposed method(5/9)
T = 3 S = [ ]
𝐶 𝑖 = {16,19,19,22,21,16,17,18,19,40,41,37,33,34,33,36} 𝐹 𝑖 = { 0,0,0,1 }
𝑊 𝑖 = {15,17,16,20,19,20,18,18,19,37,34,39,35,38,40,38} 𝑊 𝑖 ′ = { 19,20,22,23 }
𝑊 𝑖 ′ ＝ 𝑊 𝑖 +𝑏, 𝑖𝑓 𝐶 𝑖 > 𝑊 𝑖 𝑊 𝑖 −𝑏, 𝑖𝑓 𝐶 𝑖 < 𝑊 𝑖 𝑊 𝑖 ,𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
𝐶 4 = 22
𝑊 4 = 20
𝐹 4 = 1
S = [0] 𝑆 ′ = [1]
S = [00] 𝑆 ′ = [11]
S = [001] 𝑆 ′ = [110]
𝑊 𝑆 ′ ≦ 𝐶 4 + T ≦
𝑊 𝑆 ′ ≦ 𝐶 4 + T ≦
𝑊 𝑆 ′ ≦ 𝐶 4 + T ≦
𝑊 4 ′ = 23 12

13. Proposed method(6/9)
T = 3 S = [ ]
𝐶 𝑖 = {16,19,19,22,21,16,17,18,19,40,41,37,33,34,33,36} 𝐹 𝑖 = { 0,0,0,1,0,1,0,1 }
𝑊 𝑖 = {15,17,16,20,19,20,18,18,19,37,34,39,35,38,40,38} 𝑊 𝑖 ′ = { 19,20,22,23,23,15,16,18 }
𝑊 𝑖 ′ ＝ 𝑊 𝑖 +𝑏, 𝑖𝑓 𝐶 𝑖 > 𝑊 𝑖 𝑊 𝑖 −𝑏, 𝑖𝑓 𝐶 𝑖 < 𝑊 𝑖 𝑊 𝑖 ,𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
𝐶 8 = 18
𝑊 8 = 18
𝑊 8 = 𝑊 8 ′ = 18
S = [1]
𝐹 8 = 1 13

14. Proposed method(7/9)-The extracting process
𝑊 𝑖 ′ = {19,20,22,23,23,15,16,18,19,41,40,36,31,34,33,34 }
𝐹 𝑖 = { 0,0,0,1,0,1,0,1,1,1,0,1,0,1,0,0 }
I =
𝑊 𝑖 ′ 6 𝑜𝑟 7𝑏𝑖𝑡𝑠 𝐹 𝑖 1𝑏𝑖𝑡 = 𝑊 𝑖 ′′ (7 𝑜𝑟 8𝑏𝑖𝑡𝑠)
𝑊 1 ′ =19(010011) 𝐹 1 = = 𝑊 1 ′′ = (38)
𝑊 10 ′ =41( ) 𝐹 10 = = 𝑊 10 ′′ = (83)
𝑊 𝑖 ′′ = {38,40,44,47,46,31,32,37,39,83,80,73,62,69,66,68}
𝑊 𝑖 ′′ 7𝑏𝑖𝑡𝑠 𝐼 1𝑏𝑖𝑡 = 𝑊 𝑖 ′′′ (8𝑏𝑖𝑡𝑠)
𝑊 1 ′′ =38( ) 𝐼 = = 𝑊 1 ′′′ =
𝑊 𝑖 ′′′ = {76,80,89,94,92,62,64,74,79,83,80,73,62,69,66,68} 14

15. Proposed method(8/9)-The extracting process
𝑊 𝑖 ′′′ = {76,80,89,94,92,62,64,74,79,83,80,73,62,69,66,68}
𝑊 1 ′′′ = 76( ) ( ) I = (0)
𝑊 2 ′′′ = 80( ) ( ) I = (00) .
𝑊 9 ′′′ = 79( ) ( ) I = ( )
𝑊 𝑖 ′′ = {38,40,44,47,46,31,32,37,39,83,80,73,62,69,66,68}
𝑊 𝑖 ′ = {19,20,22,23,23,15,16,18,19,41,40,36,31,34,33,34 }
𝐹 𝑖 = { 0,0,0,1,0,1,0,1,1,1,0,1,0,1,0,0 } 60 70 64 80 78 75 79 74 69 77 76
Codebook
Index
𝑊 𝑖 = {15,17,16,20,19,20,18,18,19,37,34,39,35,38,40,38} 15

16. Proposed method(9/9)-The extracting process
𝑊 𝑖 = {15,17,16,20,19,20,18,18,19,37,34,39,35,38,40,38}
𝑊 𝑖 ′ = {19,20,22,23,23,15,16,18,19,41,40,36,31,34,33,34 }
𝐹 𝑖 = { 0,0,0,1,0,1,0,1,1,1,0,1,0,1,0,0 }
𝐷 ′ = | 𝑊 𝑖 ′ − 𝑊 𝑖 |
𝑊 1 ′ − 𝑊 1 =19 −15= 𝐹 1 =0 , 𝑆=100
𝑊 4 ′ − 𝑊 4 =23 −20= 𝐹 4 =1 , 𝑆=00
𝑊 8 ′ − 𝑊 8 =18 −18= 𝐹 8 =1 , 𝑆=1
S = [ ] 16

17. Experimental results(1/6)17

18. Experimental results(2/6)
PSNR (dB)
EC (bits) t
128
256
512
1024
39.76
40.51
40.85
41.11
508,571
455,625
433,687
414,215 1
39.75
40.54
40.86
508,587
455,692
433,654
414,200 2
39.77
40.53
41.09
508,523
455,563
433,681
414,207 3
39.68
40.41
40.72
41.00
516,349
464,347
441,685
421,240 4
39.39
40.06
40.40
40.68
530,220
478,194
454,103
431,922 5
38.88
39.48
39.85
40.17
545,077
492,272
466,581
442,910 6
38.24
38.80
39.19
39.56
559,125
505,526
478,164
452,784 7
37.56
38.12
38.54
38.95
571,356
516,170
487,565
460,821 8
36.76
37.32
37.75
38.20
587,778
531,407
501,852
474,221 9
35.95
36.51
36.96
37.44
599,878
542,109
510,983
482,190 10
35.08
35.62
36.09
36.56
616,456
557,886
526,264
496,980 11
34.32
34.85
35.35
35.85
627,452
567,713
534,810
504,223 12
33.51
34.01
34.49
34.99
645,617
585,524
552,235
521,444 13
32.84
33.34
33.84
34.36
655,183
593,935
559,624
527,792 14
32.15
32.63
33.11
33.62
670,523
609,070
574,836
542,747 15
31.61
32.64
33.17
678,227
615,327
580,263
547,541
 Average PSNR and EC with each threshold t. 18

19. Experimental results(3/6)19

20. Experimental results(4/6)
256
PSNR (dB)
EC (bits)
t = 0
Huang’s
Ours
Delta
Airplane
33.73
42.25
8.52
429,569
375,734
−53835
Baboon
27.55
37.16
9.61
704,569
567,010
−137559
Boat
33.21
41.47
8.26
490,188
411,482
−78706
Goldhill
34.25
42.03
7.78
552,546
441,132
−111414
Lena
34.86
42.78
7.92
454,996
380,976
−74020
Average
32.72
41.14
8.42
526,374
435,267
−91107
t = 12
34.45
0.72
511,084
81,515
33.44
5.89
676,059
−28510
34.20
0.99
548,866
58,678
34.16
−0.09
579,319
26,773
34.29
−0.57
524,093
69,097
34.11
1.39
567,884
41,511
t = 15
32.50
−1.23
539,583
110,014
31.83
4.28
704,221
−348
32.26
−0.95
579,269
89,081
32.14
−2.11
611,392
58,846
32.32
−2.54
554,907
99,911
32.21
−0.51
597,874
71,501
Comparison with Huang’s two-level encoding. 20
C.-T. Huang, M.-Y. Tsai, L.-C. Lin, W.-J. Wang, S.-J. Wang”VQ-based data hiding in IoT networks using two-level encoding with adaptive pixel replacements”J. Supercomput. (2016), pp. 1-20

21. Experimental results(5/6)21
W.-J. Wang, C.-T. Huang, S.-R. Tsuei, S.-J. Wang”A greedy steganographic SMVQ approach of greedy-USBIRDS using secret bits for image-block repairing based on differences”
Multimed. Tools Appl., 75 (22) (2016), pp. 

22. Experimental results(6/6)22
W.-J. Wang, C.-T. Huang, S.-R. Tsuei, S.-J. Wang”A greedy steganographic SMVQ approach of greedy-USBIRDS using secret bits for image-block repairing based on differences”
Multimed. Tools Appl., 75 (22) (2016), pp. 

23. Conclusions Improve embedding capacity and image quality.
dynamic-length secret bits. 23

24. -END- 24